Skip to content

Conversation

@jcf94
Copy link
Contributor

@jcf94 jcf94 commented Aug 6, 2025

Seems the migration from TVMValue to PackedArgs/AnyView missed the codegen part when I tried to test relax cutlass.
With this fix, now the codegened cutlass .cu can pass nvcc compile.

BTW, the relax/cutlass part has not been guarded by CI?

default_headers << "#include <cutlass/coord.h>\n";
default_headers << "#include <cutlass/tensor_ref.h>\n";
default_headers << "#include <cutlass/util/host_tensor.h>\n";
default_headers << "#include <tvm/runtime/logging.h>\n";
Copy link
Member

@tqchen tqchen Aug 6, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

To reduce overall dependency, assuming the codegen part did not use ICHECK or (or change to TVM_FFI_ICHECK), LOG(INFO), instead to #include <tvm/ffi/function.h> to minimize deps

@tqchen
Copy link
Member

tqchen commented Aug 6, 2025

Thanks @jcf94 ! some minor nit

@tqchen
Copy link
Member

tqchen commented Aug 8, 2025

@jcf94 would be great to do the followup minor change and see if we can bring it in. thank you

@jcf94
Copy link
Contributor Author

jcf94 commented Aug 9, 2025

@jcf94 would be great to do the followup minor change and see if we can bring it in. thank you

Sorry for just seeing your comment. 😂
I have rewritten the codegen code from CHECK to TVM_FFI_ICHECK.

@tqchen
Copy link
Member

tqchen commented Aug 9, 2025

Thanks @jcf94 I made a minor change and once it is green we can merge it in

@jcf94 jcf94 merged commit 34d4bc7 into apache:main Aug 10, 2025
12 checks passed
ShiboXing pushed a commit to ShiboXing/tvm that referenced this pull request Aug 10, 2025
* Codegen fix

---------

Co-authored-by: Tianqi Chen <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants